The article explores efficient I/O handling in programming, contrasting traditional blocking I/O with modern approaches like io_uring on Linux and kqueue on FreeBSD/macOS. In traditional I/O, reading from or writing to a file descriptor can lead to inefficiencies, especially when the data isn't immediately available, resulting in wasted time on repeated system calls. Non-blocking modes help, but they still incur high overhead due to context switches and cache misses. The shift to io_uring or kqueue allows for more efficient management of I/O by enabling the submission of readiness requests in batches, helping applications, particularly TCP servers, manage multiple connections effectively. A key feature of io_uring is its ability to let the kernel handle I/O operations directly without returning to userland after receiving readiness events. This reduces the overhead associated with system calls significantly. The article presents a simple TCP echo server example to illustrate the use of io_uring, highlighting how I/O operations can be tied to business logic. However, as applications grow, directly coupling I/O with business logic can lead to complexity. The author suggests creating a dispatch interface that abstracts the details of readiness and completion, allowing developers to schedule I/O operations with callbacks from anywhere in the code. The proposed dispatch interface can intelligently choose between io_uring and kqueue based on the kernel environment and batch requests to optimize system calls further. By using user data fields in the event structures, callbacks can be tracked and invoked upon completion without cluttering the main application logic. The article emphasizes the need for a centralized event loop and introduces a `run_for_ns` function to manage the timing of I/O operations and ensure efficient processing. This approach simplifies the programming model and reduces the risk of complicated control flow in handling I/O events.